The irrational commuter: ‘Avalanches of choice’ kill efficiency

As European cities deploy AI to manage capacity, understanding the 'stranger-following effect' is critical for the next generation of smart infrastructure.

European transit hubs are operating at a breaking point. With the global urban population projected to reach 60% by 2030, the efficiency of our train stations and airports is no longer just a matter of convenience; it is a pillar of strategic economic autonomy. Urban planners and engineers have long modeled crowd flows based on the assumption of the 'rational agent': the idea that individuals will naturally seek the fastest route from point A to point B.

New empirical data shatters this assumption. A three-year study conducted at Eindhoven Centraal station demonstrates that pedestrians frequently ignore optimal paths in favor of blindly following the person in front of them. This phenomenon, termed the 'stranger-following effect,' triggers cascading inefficiencies that threaten to choke critical infrastructure. For city planners and technology leaders, the message is clear: to solve congestion, we must stop designing for logic and start designing for human instinct.

The data behind the herd

The scale of the inefficiency is massive. Researchers from Eindhoven University of Technology (TU/e) and the University of Ferrara analyzed approximately 30 million pedestrian trajectories collected between 2021 and 2024 at Eindhoven Centraal. By narrowing this dataset to 100,000 specific passenger movements around a platform kiosk, the team isolated a distinct behavioral pattern: the 'stranger-following effect.' The data reveal that individuals in a crowd will actively choose to follow a stranger’s path, even when a visible, shorter alternative exists. This is not a random error; it is a statistically significant deviation from rational efficiency.

The study highlights that this imitation behavior is the dominant driver of collective routing choices in high-density environments. When one person makes a suboptimal choice, it does not remain an isolated incident. Instead, it triggers an 'avalanche of choice,' where subsequent pedestrians mimic the initial error, creating artificial bottlenecks in areas that should, in theory, remain clear. This finding necessitates recalibrating existing crowd dynamics models, which often fail to account for the magnitude of social imitation in transit environments.

Modeling the avalanche

Understanding the mechanics of these 'avalanches' is crucial for preventing critical failures in crowd management. The TU/e study, led by Ziqi Wang and Federico Toschi, a 2021 Ig Nobel laureate for physics, integrated this behavioral tendency into a stochastic routing model. The results show that crowd decisions are deeply rooted in local imitation behavior rather than global awareness of the environment. A single pedestrian’s decision to veer left can pull dozens of others into a congested corridor, regardless of the actual traffic flow.

This aligns with broader research in social psychology suggesting that pedestrians are not merely 'driven particles' but social beings influenced by norms and the subconscious safety of the group. In emergency scenarios, this dynamic becomes dangerous. If the 'leader' of a local cluster heads toward a blocked exit or a hazard, the 'stranger-following effect' ensures the crowd follows, potentially overriding safety signage or logic. The implications for predictive modeling are profound. Standard algorithms that assume pedestrians minimize travel cost (time and distance) are fundamentally flawed because they ignore the 'social cost' of deviating from the herd. Accurate prediction requires models that treat irrational imitation as a primary variable rather than an outlier.

AI and the predictive pivot

The industry response involves a pivot toward Artificial Intelligence (AI) that can anticipate these irrational cascades before they solidify. Cities across Europe are already deploying high-level crowd analytics. Berlin’s U-Bahn network, for instance, launched an AI-powered prediction system in 2024 that reduced peak-hour congestion by 20%. Similarly, Madrid introduced an evacuation management tool in 2025 to dynamically guide crowds during emergencies.

However, the Eindhoven findings suggest these systems must evolve. Current AI often relies on historical density and flow rate data. To be truly effective, next-generation systems must utilize computer vision to detect the specific visual cues that trigger an 'avalanche of choice'. By identifying the moment a crowd begins to lock onto a suboptimal leader, AI-driven control systems can intervene. This intervention could take the form of dynamic digital signage that changes in real-time to break the line of sight or mobile alerts that push users toward underutilized exits. The goal is to use technology to disrupt the psychological feedback loop of imitation, effectively 'nudging' the herd back toward efficiency without human intervention.

Engineering against instinct

Beyond digital intervention, physical infrastructure must be redesigned to mitigate the biological impulse to follow. Research into self-organized crowd dynamics suggests that geometric boundary conditions play a pivotal role in stabilizing flow. If wide-open concourses encourage herd behavior, introducing strategic obstacles can force individuals to disengage from the person ahead and re-evaluate their path. 'Stripe formation', a phenomenon in which opposing flows naturally organize into lanes, can be induced by the placement of columns or zigzag-shaped barriers. These architectural interventions increase cognitive load slightly, forcing pedestrians to look up and assess the environment rather than stare at the back of a stranger’s coat. In high-density venues such as stadia or airports, increasing the diameter of egress routes is a standard solution, but the TU/e findings suggest that visibility is equally important. If pedestrians can clearly see a destination rather than just the crowd, the reliance on stranger-following diminishes. Future infrastructure projects must prioritize line-of-sight engineering to counteract the natural tendency toward 'avalanches of choice'.

The surveillance dilemma

Implementing these solutions requires a dense network of sensors, cameras, and data processing, raising significant ethical and legal questions. The Eindhoven study itself relied on overhead sensors to track anonymous trajectories, a method that attempts to balance data utility with privacy. However, as cities scale these technologies, the line between crowd management and mass surveillance blurs. The European Union’s AI Act and GDPR impose strict limitations on biometric identification and data retention, creating a compliance minefield for transit operators.

The challenge lies in obtaining informed consent from millions of transient commuters, a practical impossibility in a busy train station. Furthermore, relying on AI to direct human movement touches on issues of individual autonomy. If an algorithm steers a crowd to a specific exit to optimize system flow, it may prioritize the collective over individual preferences or speed. As Europe pushes for strategic autonomy in technology and infrastructure, maintaining public trust is paramount. Operators must ensure that the 'smart station' does not become a panopticon, using non-invasive LiDAR and anonymized data processing to manage the herd without tracking the individual.